def test_object():
foo = Object(bar='bar',
baz=(1, 2, 3),
foo1=Object(bar='BAR', baz={
'one': 1,
'two': 2
}))
assert foo.bar == 'bar'
assert foo.baz == (1, 2, 3)
assert foo.foo1.bar == 'BAR'
assert foo.foo1.baz['one'] == 1
def get_predict_v(self, x, score):
# [N]
item_pop = tf.cast(self.data.item_pop, tf.float32)
item_pop_log = tf.log(item_pop + np.e)
item_deg_self = tf.cast(tf.gather(self.data.item_deg_self_per_phase, x.phase), tf.float32)
item_pop_self_log = tf.log(item_deg_self + np.e)
if args.mode_pop == 'log':
score = score * args.alpha_pop_base + score / item_pop_log
elif args.mode_pop == 'log_mdeg':
score = score * item_pop_self_log / item_pop_log
elif args.mode_pop == 'log_mdeg_only':
score = score * item_pop_self_log
elif args.mode_pop == 'linear':
item_pop = tf.cast(self.data.item_pop, tf.float32) + 1.0
score = score * args.alpha_pop_base + score / item_pop
elif args.mode_pop == 'log_md':
item_pop_self_log = tf.log(item_deg_self + np.e + 10.0)
score = score * item_pop_self_log / item_pop_log
if args.mode_rare in {'log', 'linear', 'base'}:
if args.mode_rare == 'log':
rare_weight_pop = 1.0 / tf.log(tf.cast(self.data.item_pop, tf.float32) + np.e)
elif args.mode_rare == 'linear':
rare_weight_pop = 1.0 / tf.cast(self.data.item_pop, tf.float32)
elif args.mode_rare == 'base':
rare_weight_pop = 0.0
else:
raise Exception
# [N]
rare_weight = float(args.alpha_rare)
rare_weight = rare_weight + rare_weight_pop
rare_weight *= float(args.alpha_rare_mul)
is_rare = self.is_rare(x)
score = tf.where(is_rare, score * rare_weight + float(args.alpha_rare_base), score)
score = UTILS.mask_logits(score, x.score_mask)
tf.summary.histogram('score', score)
top_items, top_scores = self.topk_idx(score, x)
if args.dump_all:
self.tmp_vars.update(all_scores=score, item_seq=x.seq, ts_seq=x.ts, q_ts=x.q_ts)
ret = Object(user=x.user, phase=x.phase, top_items=top_items, top_scores=top_scores)
ret.update(**self.tmp_vars)
return ret
return Object(user=x.user, phase=x.phase, top_items=top_items, top_scores=top_scores)
def test_story_runner_returns_a_result_with_the_original_Fixture():
mocker = Mocker()
settings = Settings()
fixture = Fixture()
action = some_action()
fixture.append_story(action.scenario.story)
runner = StoryRunner()
context = Object()
context.browser_driver = mocker.mock()
context.browser_driver.start_test("http://localhost")
context.browser_driver.stop_test()
context.settings = mocker.mock()
context.settings.on_before_action
mocker.result(None)
context.settings.on_action_successful
mocker.result(None)
context.language = mocker.mock()
context.language.get('given')
mocker.result('Given')
context.language.get('when')
mocker.result('When')
context.language.get('then')
mocker.result('Then')
with mocker:
result = runner.run_stories(settings, fixture, context=context)
assert result.fixture == fixture
def test_should_catch_assertion_error():
mocker = Mocker()
def action_failed_method(context, *args, **kwargs):
assert False, "bla"
settings = Settings()
runner = StoryRunner()
fixture = Fixture()
action = some_action()
fixture.append_story(action.scenario.story)
action.execute_function = action_failed_method
context = Object()
context.browser_driver = mocker.mock()
context.browser_driver.start_test("http://localhost")
context.browser_driver.stop_test()
context.settings = mocker.mock()
context.settings.on_before_action
mocker.result(None)
context.settings.on_action_error
mocker.result(None)
context.language = mocker.mock()
context.language.get('given')
mocker.result('Given')
with mocker:
result = runner.run_stories(settings=settings,
fixture=fixture,
context=context)
assert isinstance(action.error, AssertionError)
assert action.error.message == "bla"
def test_should_handle_action_errors_successfully():
mocker = Mocker()
def action_failed_method(context, *args, **kwargs):
raise ActionFailedError("bla")
settings = Settings()
runner = StoryRunner()
fixture = Fixture()
action = some_action()
fixture.append_story(action.scenario.story)
action.execute_function = action_failed_method
context = Object()
context.browser_driver = mocker.mock()
context.browser_driver.start_test("http://localhost")
context.browser_driver.stop_test()
context.settings = mocker.mock()
context.settings.on_before_action
mocker.result(None)
context.settings.on_action_error
mocker.result(None)
context.language = mocker.mock()
context.language.get('given')
mocker.result('Given')
with mocker:
result = runner.run_stories(settings=settings,
fixture=fixture,
context=context)
assert fixture.get_status() == Status.Failed
def test_should_execute_scenarios_successfully():
mocker = Mocker()
settings = Settings()
runner = StoryRunner()
fixture = Fixture()
fixture.append_story(some_action().scenario.story)
context = Object()
context.browser_driver = mocker.mock()
context.browser_driver.start_test("http://localhost")
context.browser_driver.stop_test()
context.settings = mocker.mock()
context.settings.on_before_action
mocker.result(None)
context.settings.on_action_successful
mocker.result(None)
context.language = mocker.mock()
context.language.get('given')
mocker.result('Given')
context.language.get('when')
mocker.result('When')
context.language.get('then')
mocker.result('Then')
with mocker:
result = runner.run_stories(settings=settings,
fixture=fixture,
context=context)
assert fixture.get_status() == Status.Successful
def forward(self, x):
# user: [BS,], seq: [BS, L], ts: [BS, L]
# [BS, k]
seq_rep = self.get_out_rep(x)
if args.use_bl:
seq_rep = tf.layers.dense(seq_rep,
args.dim_k,
name='bl',
kernel_regularizer=UTILS.l2_loss('bl'),
use_bias=False)
dropout = tf.where(self.is_on_train, args.dropout, 1.0)
seq_rep = tf.nn.dropout(seq_rep, dropout)
score = self.calculate_score_for_pred(x, seq_rep, self.item_emb_matrix)
# [BS, M]
neg = self.sample_neg(x.ans, score)
if args.mode_loss == 'softmax':
loss = self.get_loss_softmax(x, seq_rep, neg)
elif args.mode_loss == 'pairwise':
loss = self.get_loss_pairwise(x, seq_rep, neg)
else:
a = float(args.mode_loss)
loss_1 = self.get_loss_softmax(x, seq_rep, neg)
loss_2 = self.get_loss_pairwise(x, seq_rep, neg)
loss = loss_1 * a + loss_2 * (1.0 - a)
tf.summary.scalar('loss', loss)
train_op = UTILS.minimizer(loss)
train_v = Object(loss=loss)
return train_op, train_v, self.get_predict_v(x, score)
def test_execute_action_will_not_execute_itself():
mocker = Mocker()
class DoSomethingRecursiveAction(ActionBase):
regex = r'^(And )?I do "(?P<what>\w+)" stuff$'
def execute(self, context, getter_mock, *args, **kwargs):
self.execute_action('And I do "recursive" stuff',
context,
getter=getter_mock)
language_getter_mock = mocker.mock()
language_getter_mock.get(ANY)
mocker.count(min=1, max=None)
mocker.result('^$')
context_mock = Object(settings=mocker.mock())
context_mock.settings.default_culture
mocker.result("en-us")
with mocker:
dosaction = DoSomethingRecursiveAction()
args = []
kwargs = dict(what='nothing')
dosaction.execute(context_mock,
getter_mock=language_getter_mock,
*args,
**kwargs)
def standardize_new(xt, yt, xv, yv, xte, yte, is_regression):
ss = Object()
xt, xte, ss.mean_x_train, ss.std_x_train = standardize(xt, xte)
if xv is not None:
xv = (xv - ss.mean_x_train[None, :]) / ss.std_x_train[None, :]
if is_regression:
yt, yte, ss.mean_y_train, ss.std_y_train = standardize(yt, yte)
if yv is not None:
yv = (yv - ss.mean_y_train[None]) / ss.std_y_train[None]
return xt, yt, xv, yv, xte, yte, ss
def getInfo(self):
o = Object()
for prop, ctype in self.elaCarrier.contents._fields_:
# should check for pointers
try:
o[prop] = getattr(self.elaCarrier.contents, prop)
except Exception as e:
print e
o[prop] = ctype
return o
def __init__(self, data: dataset.Data):
self.tmp_vars = Object()
self.data = data
# self.save_dir = f'{utils.save_dir}/{args.run_name}'
self.save_dir = f'{utils.save_dir}/{args.msg}'
with self.data.tf_graph.as_default():
tf.set_random_seed(args.seed)
self.compile()
self.fit_step = 0
def make_io(self):
self.is_on_train = tf.placeholder(tf.bool, [], 'is_on_train')
train_data = self.data.train_batch_repeat
train_data_iter = train_data.make_one_shot_iterator()
self.train_data_handle = self.sess.run(train_data_iter.string_handle())
self.data_handle = tf.placeholder(tf.string, [], 'data_handle')
data_iter = tf.data.Iterator.from_string_handle(
self.data_handle,
train_data.output_types,
train_data.output_shapes,
)
self.input_dict = data_iter.get_next()
self.input_dict = Object(**self.input_dict)
def test_story_runner_returns_a_result():
mocker = Mocker()
settings = Settings()
fixture = Fixture()
runner = StoryRunner()
context = Object()
context.browser_driver = mocker.mock()
context.browser_driver.start_test("http://localhost")
context.browser_driver.stop_test()
with mocker:
result = runner.run_stories(settings, fixture, context=context)
assert result is not None
def make_context_and_fso_mocks(mocker):
hooks_dir = ["/hooks/dir/"]
pages_dir = ["/pages/dir/"]
custom_actions_dir = ["/custom/actions/dir/"]
context_mock = Object()
context_mock.browser_driver = mocker.mock()
context_mock.settings = mocker.mock()
context_mock.settings.hooks_dir
mocker.count(min=1, max=None)
mocker.result(hooks_dir)
context_mock.settings.pages_dir
mocker.count(min=1, max=None)
mocker.result(pages_dir)
context_mock.settings.custom_actions_dir
mocker.count(min=1, max=None)
mocker.result(custom_actions_dir)
context_mock.settings.base_url
mocker.count(min=0, max=None)
mocker.result("http://localhost")
context_mock.settings.default_culture
mocker.count(min=1, max=None)
mocker.result("en-us")
files = ["/some/weird/file.py"]
fso_mock = mocker.mock()
fso_mock.add_to_import(hooks_dir[0])
fso_mock.add_to_import(pages_dir[0])
fso_mock.add_to_import(custom_actions_dir[0])
fso_mock.locate(hooks_dir[0], '*.py')
mocker.result(files)
fso_mock.locate(pages_dir[0], '*.py')
mocker.result(files)
fso_mock.locate(custom_actions_dir[0], '*.py')
mocker.result(files)
fso_mock.import_file(ANY)
mocker.count(min=1, max=None)
fso_mock.remove_from_import(custom_actions_dir[0])
mocker.count(min=1, max=None)
fso_mock.remove_from_import(pages_dir[0])
mocker.count(min=1, max=None)
fso_mock.remove_from_import(hooks_dir[0])
mocker.count(min=1, max=None)
return context_mock, fso_mock
def test_action_base_can_resolve_elements_in_a_given_page():
mocker = Mocker()
class DoOtherThingAction(ActionBase):
regex = "^Do other thing$"
def execute(self, context, *args, **kwargs):
self.element = self.resolve_element_key(context, "button",
"Something")
context_mock = Object(current_page=mocker.mock())
context_mock.current_page.get_registered_element("Something")
mocker.result("btnSomething")
with mocker:
action = DoOtherThingAction()
action.execute(context_mock)
assert action.element == "btnSomething"
def test_action_base_can_resolve_elements_using_browser_driver():
mocker = Mocker()
class DoOneMoreThingAction(ActionBase):
regex = "^Do other thing$"
def execute(self, context, *args, **kwargs):
self.element = self.resolve_element_key(context, "button",
"Something")
context_mock = Object(browser_driver=mocker.mock(), current_page=None)
context_mock.browser_driver.resolve_element_key(context_mock, "button",
"Something")
mocker.result("btnSomething")
with mocker:
action = DoOneMoreThingAction()
action.execute(context_mock)
assert action.element == "btnSomething"
def get_metric_v(self, x, predict_v):
# [BS,]
true_item = x.ans
true_item_a1 = tf.expand_dims(true_item, -1)
# [BS, M], [BS, 1]
eq = tf.cast(tf.equal(predict_v.top_items, true_item_a1), tf.int32)
# [BS,]
m = tf.reduce_max(eq, -1)
idx = tf.cast(tf.argmax(eq, -1), tf.int32)
rank = idx + m - 1
ndcg = tf.log(2.0) * tf.cast(m, tf.float32) / tf.log(2.0 + tf.cast(idx, tf.float32))
hit_rate = tf.cast(m, tf.float32)
ret = Object(
ndcg=ndcg,
hit_rate=hit_rate,
user=x.user,
true_item=true_item,
phase=x.phase,
top_items=predict_v.top_items,
top_scores=predict_v.top_scores,
rank=rank,
q_ts=x.q_ts,
)
return ret
def __init__(self, carrier=None):
self.events = Object()
self.elaCallbacks = ELACALLBACKS()
self.events.idle = []
self.elaCallbacks.idle = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_void_p)(self.__idleCB__)
self.events.connectionStatus = []
self.elaCallbacks.connection_status = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ELACONNECTIONSTATUS,
ctypes.c_void_p)(self.__connectionStatusCB__)
self.events.ready = []
self.elaCallbacks.ready = ctypes.CFUNCTYPE(None,
ctypes.POINTER(ELACARRIER),
ctypes.c_void_p)(
self.__readyCB__)
self.events.selfInfo = []
self.elaCallbacks.self_info = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.POINTER(ELAUSERINFO),
ctypes.c_void_p)(self.__selfInfoCB__)
self.events.friendList = []
self.elaCallbacks.friend_list = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.POINTER(ELAFRIENDINFO),
ctypes.c_void_p)(self.__friendListCB__)
self.events.friendConnection = []
self.elaCallbacks.friend_connection = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p,
ELACONNECTIONSTATUS, ctypes.c_void_p)(self.__friendConnectionCB__)
self.events.friendInfo = []
self.elaCallbacks.friend_info = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p,
ctypes.POINTER(ELAFRIENDINFO),
ctypes.c_void_p)(self.__friendInfoCB__)
self.events.friendPresence = []
self.elaCallbacks.friend_presence = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p,
ELAPRESENCESTATUS, ctypes.c_void_p)(self.__friendPresenceCB__)
self.events.friendRequest = []
self.elaCallbacks.friend_request = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p,
ctypes.POINTER(ELAUSERINFO), ctypes.c_char_p,
ctypes.c_void_p)(self.__friendRequestCB__)
self.events.friendAdded = []
self.elaCallbacks.friend_added = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.POINTER(ELAUSERINFO),
ctypes.c_void_p)(self.__friendAddedCB__)
self.events.friendRemoved = []
self.elaCallbacks.friend_removed = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p,
ctypes.c_void_p)(self.__friendRemovedCB__)
self.events.friendMessage = []
self.elaCallbacks.friend_message = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p, ctypes.c_char_p,
ctypes.c_size_t, ctypes.c_void_p)(self.__friendMessageCB__)
self.events.friendInvite = []
self.elaCallbacks.friend_invite = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p, ctypes.c_char_p,
ctypes.c_char_p, ctypes.c_size_t,
ctypes.c_void_p)(self.__friendInviteCB__)
self.events.groupInvite = []
self.elaCallbacks.group_invite = ctypes.CFUNCTYPE(
None, ctypes.POINTER(ELACARRIER), ctypes.c_char_p, ctypes.c_char_p,
ctypes.c_size_t, ctypes.c_void_p)(self.__groupInviteCB__)
self.events.friendsIterate = []
self.friendsIterate = ELAFRIENDSITERATECALLBACK(
self.__friendsIterateCB__)
self.events.friendInviteResponse = []
self.friendInviteResponse = ELAFRIENDINVITERESPONSECALLBACK(
self.__friendInviteResponseCB__)
class Base:
args = Object()
need_train = True
def __init__(self, data: dataset.Data):
self.tmp_vars = Object()
self.data = data
# self.save_dir = f'{utils.save_dir}/{args.run_name}'
self.save_dir = f'{utils.save_dir}/{args.msg}'
with self.data.tf_graph.as_default():
tf.set_random_seed(args.seed)
self.compile()
self.fit_step = 0
def compile(self):
self.emb_l2_norm_op = None
self.sess = UTILS.get_session()
self.make_io()
self.make_model()
self.sess.run(tf.global_variables_initializer())
if self.emb_l2_norm_op is not None:
self.sess.run(self.emb_l2_norm_op)
def make_io(self):
self.is_on_train = tf.placeholder(tf.bool, [], 'is_on_train')
train_data = self.data.train_batch_repeat
train_data_iter = train_data.make_one_shot_iterator()
self.train_data_handle = self.sess.run(train_data_iter.string_handle())
self.data_handle = tf.placeholder(tf.string, [], 'data_handle')
data_iter = tf.data.Iterator.from_string_handle(
self.data_handle,
train_data.output_types,
train_data.output_shapes,
)
self.input_dict = data_iter.get_next()
self.input_dict = Object(**self.input_dict)
def get_metric_v(self, x, predict_v):
# [BS,]
true_item = x.ans
true_item_a1 = tf.expand_dims(true_item, -1)
# [BS, M], [BS, 1]
eq = tf.cast(tf.equal(predict_v.top_items, true_item_a1), tf.int32)
# [BS,]
m = tf.reduce_max(eq, -1)
idx = tf.cast(tf.argmax(eq, -1), tf.int32)
rank = idx + m - 1
ndcg = tf.log(2.0) * tf.cast(m, tf.float32) / tf.log(2.0 + tf.cast(idx, tf.float32))
hit_rate = tf.cast(m, tf.float32)
ret = Object(
ndcg=ndcg,
hit_rate=hit_rate,
user=x.user,
true_item=true_item,
phase=x.phase,
top_items=predict_v.top_items,
top_scores=predict_v.top_scores,
rank=rank,
q_ts=x.q_ts,
)
return ret
def get_predict_v(self, x, score):
# [N]
item_pop = tf.cast(self.data.item_pop, tf.float32)
item_pop_log = tf.log(item_pop + np.e)
item_deg_self = tf.cast(tf.gather(self.data.item_deg_self_per_phase, x.phase), tf.float32)
item_pop_self_log = tf.log(item_deg_self + np.e)
if args.mode_pop == 'log':
score = score * args.alpha_pop_base + score / item_pop_log
elif args.mode_pop == 'log_mdeg':
score = score * item_pop_self_log / item_pop_log
elif args.mode_pop == 'log_mdeg_only':
score = score * item_pop_self_log
elif args.mode_pop == 'linear':
item_pop = tf.cast(self.data.item_pop, tf.float32) + 1.0
score = score * args.alpha_pop_base + score / item_pop
elif args.mode_pop == 'log_md':
item_pop_self_log = tf.log(item_deg_self + np.e + 10.0)
score = score * item_pop_self_log / item_pop_log
if args.mode_rare in {'log', 'linear', 'base'}:
if args.mode_rare == 'log':
rare_weight_pop = 1.0 / tf.log(tf.cast(self.data.item_pop, tf.float32) + np.e)
elif args.mode_rare == 'linear':
rare_weight_pop = 1.0 / tf.cast(self.data.item_pop, tf.float32)
elif args.mode_rare == 'base':
rare_weight_pop = 0.0
else:
raise Exception
# [N]
rare_weight = float(args.alpha_rare)
rare_weight = rare_weight + rare_weight_pop
rare_weight *= float(args.alpha_rare_mul)
is_rare = self.is_rare(x)
score = tf.where(is_rare, score * rare_weight + float(args.alpha_rare_base), score)
score = UTILS.mask_logits(score, x.score_mask)
tf.summary.histogram('score', score)
top_items, top_scores = self.topk_idx(score, x)
if args.dump_all:
self.tmp_vars.update(all_scores=score, item_seq=x.seq, ts_seq=x.ts, q_ts=x.q_ts)
ret = Object(user=x.user, phase=x.phase, top_items=top_items, top_scores=top_scores)
ret.update(**self.tmp_vars)
return ret
return Object(user=x.user, phase=x.phase, top_items=top_items, top_scores=top_scores)
def make_model(self):
with tf.variable_scope('Network', reuse=tf.AUTO_REUSE, regularizer=UTILS.l2_loss('all')):
x = self.input_dict
self.train_op, self.train_v, self.predict_v = self.forward(x)
self.metric_v = self.get_metric_v(x, self.predict_v)
self.metric_v.update(loss=self.train_v.loss)
network_var_list = tf.trainable_variables(scope='^Network/')
if network_var_list:
args.log.log('trainable_variables:')
for v in network_var_list:
args.log.log(f'network: {v}')
tf.summary.histogram(v.name, v)
self.saver = tf.train.Saver(var_list=tf.trainable_variables())
# self.saver_emb = tf.train.Saver(var_list=tf.trainable_variables(scope='^Network/Emb_'))
def fit(self):
data = {
self.is_on_train: True,
self.data_handle: self.train_data_handle,
}
tb_v = []
if args.run_tb:
tb_v = [self.all_summary]
debug_v = DEBUG.fit_show_list
all_v = [self.train_op, self.train_v, debug_v, tb_v]
_, train_v, debug_v, tb_v = self.sess.run(all_v, data)
if self.emb_l2_norm_op is not None:
self.sess.run(self.emb_l2_norm_op)
if tb_v:
self.tbfw.add_summary(tb_v[0], self.fit_step)
DEBUG.when_run(debug_v)
self.fit_step += 1
return train_v
def inference(self, data, out_obj):
with self.data.tf_graph.as_default():
data_iter = data.make_one_shot_iterator()
data_handle = self.sess.run(data_iter.string_handle())
data = {
self.is_on_train: False,
self.data_handle: data_handle,
}
while True:
try:
ret_value, debug_v = self.sess.run([out_obj, DEBUG.inf_show_list], data)
DEBUG.when_run(debug_v)
yield ret_value
except tf.errors.OutOfRangeError:
break
def metric(self, data):
for v in self.inference(data, self.metric_v):
yield v
def predict(self, data):
for v in self.inference(data, self.predict_v):
yield v
def save(self, s):
if not self.need_train:
return
name = f'{self.save_dir}/model_{s}.ckpt'
self.saver.save(self.sess, name)
def restore(self, s):
if not self.need_train:
return
name = f'{self.save_dir}/model_{s}.ckpt'
self.saver.restore(self.sess, name)
def restore_from_other(self, run_name):
save_dir = f'{utils.save_dir}/{run_name}'
s = 0
if not self.need_train:
return
import os
if not os.path.isdir(save_dir):
args.log.log('download from hdfs')
sh = f'$HADOOP_HOME/bin/hadoop fs -get save/{utils.project_name}/{run_name} {utils.save_dir}/'
print(os.system(sh))
name = f'{save_dir}/model_{s}.ckpt'
self.saver.restore(self.sess, name)
# if args.restore_train:
# self.saver_emb.restore(self.sess, name)
# else:
# self.saver.restore(self.sess, name)
def forward(self, x):
raise NotImplementedError
def is_rare(self, x):
is_rare = tf.gather(self.data.is_rare_per_phase, x.phase)
return is_rare
def topk_idx(self, prob, x):
rare_k = args.nb_rare_k
if rare_k < 0:
topk = tf.nn.top_k(prob, args.nb_topk)
return topk.indices, topk.values
is_rare = self.is_rare(x)
prob_rare = UTILS.mask_logits(prob, is_rare)
prob_rich = UTILS.mask_logits(prob, tf.logical_not(is_rare))
topk_rare = tf.nn.top_k(prob_rare, rare_k).indices
topk_rich = tf.nn.top_k(prob_rich, args.nb_topk - rare_k).indices
topk = tf.concat([topk_rich, topk_rare], -1)
# [BS, N], [BS, L] --> [BS, L]
top_prob = tf.batch_gather(prob, topk)
sort_topk = tf.nn.top_k(top_prob, args.nb_topk)
sort_idx = sort_topk.indices
top_values = sort_topk.values
sorted_topk = tf.batch_gather(topk, sort_idx)
return sorted_topk, top_values
def before_train(self):
pass
def after_train(self):
pass
"""Configurations."""
from utils import Object, color
infinite_generation = True
map_params = Object(
unit_size=10,
height=20,
cell_size=3,
wall_width=1,
field_size=15,
core_size=3,
center_size=3,
seed=10,
colors=Object(
wall=(1, 0.5, 0.5, 1),
floor=color("94DEFF"),
sky=color("FFF9D9")
)
)
key_map = Object(
character=Object(
forward="w",
backward="s",
left="a",
right="d"
),
utility=Object(
pause="escape"
)
